JPS63142478A - White line detector - Google Patents

Abstract

PURPOSE:To stably detect a white line by detecting a white line candidate part and setting up a threshold in a maximum variable density value on a parameter plane obtained by HOUGH conversion. CONSTITUTION:A point changing a variable density value from dark to bright and a point changing the variable density value from bright to dark are detected from a two-dimensional observation image obtained by image-picking up external environment including a white line by an edge detecting means 100 and an edge formed by the assembly of the detected points is detected. When the area formed by the edge is less than a prescribed width, the area is detected as a white line candidate part by a white line candidate part detecting means 200. A pattern sorting means 300 classifies the pattern of the white line and a white line detecting means 400 executes the HOUGH conversion of the classified pattern to detect the white line.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a white line detection device for detecting white lines drawn on a travel path of an automatic guided vehicle or the like.

(Prior art) In order to control the running of a traveling vehicle such as an automatic guided vehicle, the external environment is imaged by an imaging unit such as a television camera attached to the traveling vehicle, and the external environment is recognized from the captured two-dimensional image. There are known techniques to do this. Conventionally, the first step in recognizing the external environment for this purpose is to use line-shaped or spot-shaped landmarks that can be sensed on the road surface or wall surface.
A known method is to set up markers (markers) and detect the markers using a light-receiving element provided on a vehicle (for example, Japanese Patent Laid-Open No. 59-224505).

However, in this conventional method, special markers must be installed on the road or wall surface, which makes it difficult to change the layout of the external environment, and the markers are susceptible to dirt. It is desired to develop a method to stably recognize the external environment without making any changes to the external environment.

(Purpose and configuration of IJ1) The present invention was made in consideration of the point L, and it is possible to recognize the external environment without making any changes to the external environment such as reinstalling markers when changing the layout or expanding the installation. For this purpose, we focused on the fact that white lines are drawn inside the factory to distinguish the paths for people and vehicles from the workplace, and in order to achieve the above purpose, the overall configuration is shown in Figure 1. As shown, from the two-dimensional observation image obtained by imaging the external environment including the white line, the edge detection means lOO detects points where the gray value changes from dark to light and points where the gray value changes from bright to dark. The white line candidate detecting means 200 detects an edge formed by a set of edges, and the white line candidate detecting means 200 detects an area where the edge formed by the edge is less than a predetermined width as a white line candidate, and the pattern classifying means 300 detects the area The white line detection means 40G classifies the white line pattern from the set of center points of
The system is configured to detect white lines by performing conversion.

(Example) The present invention will be explained below based on the drawings.

FIG. 2 shows a block diagram of an embodiment of the white line detection device according to the present invention.

This device includes an imaging unit 101, an image memory 102 that stores image data captured by the imaging unit ioi, and a differential calculation unit 103 that calculates edge information from the information stored in the image memory 102. Binarize the edge information with a threshold α 2 (Extracted by an i conversion unit 104, a binarization unit 105 that inverts and binarizes the calculated edge information with a threshold β, and these binarization units 104 and 105) Line thinning units 106 and 1 thin the edge information into a line 1, respectively.
07, and an adding unit 10 that adds the two thinned images.
8, a vertical edge removal unit 109 that removes vertical edges from the addition result, a white line candidate extraction unit 110 that detects a region that is considered to be a white line from there, and a center line (skeletal line) from the area that is considered a white line candidate. a skeletal line extraction unit itt that extracts the skeletal lines, a pattern classification unit 112 that divides the extracted skeletal lines into patterns based on whether they are upward-right or upward-left, and an image memory for storing upward-right skeleton lines that stores the obtained patterns. 1
13, a similar image memory 114 for storing right-down skeleton lines, and a white line detection unit 115 that detects white lines from data in these memories 113 and 114.

Next, the operation will be explained with reference to FIG.

An image signal A(i, j) as shown in FIG. An edge image A'(i,j) is obtained.

For example, when the image to be handled can have 8-bit gradation information (gradation value 0 to 255) per pixel, A'(i, j)=
lNT[(A(i+l,j-1)+2A(i+l,
j)+A(isu1゜j◆1)-A(i-1,j-1)-
2A(i-1,j)-A(i-1,j+1)+1024
)/23]. Here, INT(X) is a relationship that gives the maximum integer that does not exceed X, and both 1024 and 23 are terms for normalization.

The edge image A'(i, j) obtained in this way has two images i
A'(t T J ) is sent to the converting units 104 and 105, and the binarizing unit 104 binarizes A'(t T J ) with a certain threshold value exceeding the gray level value 128. For pixels that exceed
55, and replace the gray value 0 for pixels that do not exceed the threshold), the gray value 2 is the point where the gray changes from bright to dark above a certain value during raster scanning.
55, and the binarization unit 105 inverts and binarizes A''(i, j) with a certain threshold that does not exceed the gradation value 12B. For pixels whose value exceeds the threshold, the gray value is replaced with 0, and for pixels whose value does not exceed the threshold, the gray value is replaced with 12B), from brightness of more than a certain value during raster scanning. A point where the gray level changes to dark is detected as a pixel having a gray level value of 255.

Next, the image obtained in the 2-1 conversion units 104 and 105 is thinned to a thickness l in the line thinning units 106 and 107, and the line thinning unit 107 produces a shading of 128 at the obtained shading change point from bright to dark. The value is given to the thinning unit 106 in the adding unit 108
Then, the grayscale values of the same pixels resulting from the thinning unit 107 are added. Through the above processing, for example, as shown in Fig. 3 (a)
An edge constituent point image B (i, j) as shown in FIG.

Next, the vertical edge removal unit 109 removes the vertical run with a gray value of 255 and the vertical run with a gray value of 128 in the edge constituent point image B (i, j), resulting in the image shown in FIG. An image C(i,j) like this is obtained.

Next, the white line candidate extracting unit 110 checks the shading values of the vertical edge removed image C(i, j), and determines that the shading value of the left end is 25.
5. In a horizontal area where the shading value at the right end is 128, the width X at both ends is set in advance and W is given in Figure 3 (d), so that W/2≦X≦W is satisfied. We searched for a region with a shading value of 255 as a white line candidate, and
An image D (i, j) as shown in e) is obtained.

FIG. 4 is a flowchart of white line candidate extraction. Vertical M
Vertical edge removed image C (i, j
), i=1. With j=1 (F-1), raster scan is performed in the horizontal direction from the upper left of the screen. During this time, image C (
Check the shading values of i, j) and check the point where the shading value becomes 255 (F-2). When a pixel with an image density value of 255 is found, a horizontal search range W determined by the following equation is set (F-3).

W=INT (j/20) Here, 20 in the formula is an example of a value that is predetermined by the user based on the camera layout, the thickness of the white line, etc.

, Next, confirm that W is not 0 (F-4), and once confirmed, calculate K from =i+W (F-5), and find that ≧N
(F-6) The gray value of image C (i, j) is 12 until (F-6)
Check the point that becomes 8 (F-7). When the gradation value reaches 128, set L=i and F-8), D (L, j)=2
Find the pixel that is 55 (F-9). After this process, return to step (F-2), i=1 to N, j=1 to M
Repeat the process described above for . In this way, the screen is searched from left to right, and the largest area not exceeding W is extracted from the image C(i, j), resulting in a white line candidate image D(i, j) as shown in FIG. j) is obtained.

That is, in this process as well, it is assumed that the screen is first scanned in the horizontal direction starting from the upper left pixel, and i=1 and j=1 (P-1).

First, it is checked whether the gradation value of the white line candidate portion image D (i, j) is 0 or not (P-2). If it is O, it is assumed that there is no white line candidate portion, and the white line candidate portion image D (i, j) is moved by one pixel in the K-plane direction.

On the other hand, if the gradation value of image D (i, j) is not 0, it is assumed that there is a white line candidate and A=1 (p-3), then i
as i+1 (P-4) until i=N (P-5)
, continues to check whether the density value of image D (i, j) is O or not (
P-6). Each time the scanning point moves horizontally by one pixel, A is incremented by l (F-7).

In step (P-6), find D(i, j)=O (P-8), and calculate the pixel E(k,
Find j)=255 (P-9).

Returning to step (P-2), the above-described process is repeated for i=1 to N and j=1 to M to obtain skeleton lines for the entire screen. As a result, a skeleton line image E (i, j) as shown in FIG. 3(f) is obtained.

Next, in the pattern separation unit 112, the skeleton line image E (
A skeleton line F (i, j) upward to the right and a skeleton line G (i, j) downward to the right are extracted from i, j). That is, E(
If you apply a 3 x 3 mask to i, j) so that the skeleton line constituent pixels (pixels with an e-light value of 255) are placed in the center, the pattern within the mask will be one of the batanine patterns shown in Figure 6. , right-up pattern, and right-down pattern can be easily classified. The upper right skeleton line F (
i, j) and the right-downward skeleton 1iG(i, j) as shown in FIGS. 3(g) and 3(h), and are stored in image memories 113 and 114, respectively.

Next, the white line detection unit 115 detects these skeleton lines F (
A white line can be detected by applying HOUGH transformation to each of i, j) and G(i, j). The white line detection results are shown in Figure 3. Figure 1. and the white line detected by fLR.

HOUGH transformation is a method for detecting straight lines from a sequence of points such as F (i, j) and G (i, j), and is widely used in the field of image processing. Hereinafter, the HOUGH transformation used this time will be explained using FIG. 7, taking F (i, j) as an example.

Figure 7 (b) shows the reference line y = α set on the same figure (a).
This figure represents the locus of the point Q where the straight line connecting one point P and the skeleton constituent point A intersects with the edge of the image when the point P is changed from Po to Pn. A point on the parameter plane (Pl, Q
t), the gradation value is incremented each time the trajectory passes in order to represent how many skeleton line constituent points are on the line segment connecting Pt and Qt. For example, in Figure 7 (b),
A gradation value 2 is given to the intersection of the projection pattern 1 of point A and the projection pattern sentence of point B, and a gradation value 1 is given to the other points on l s and la.

Figure 8 shows a flowchart of the HOUGH transformation of F (i, j), in which the gradation values of the parameter plane are converted to R(P,
Q).

First, set the point P to 20 (W-1), and start from the leftmost pixel (1 = l # J = α) on the reference line (W-
2) Increase i from 1 to P and j from α to M until the gray value of the upward-right skeleton line F (i, j) reaches 255. When the shading value of the upper right skeleton line F (i, j) becomes 255, calculate Ql (W-4), and increment the shading value R (P, Q) of the parameter plane by l (
W-S). Here, the maximum shading value of the parameter plane V = 0,
x=p,,Y''Q Starting from o (W-6), change X from Po to Pn, Y from Q. to Qn, and reach the point (x.

Yl)e Find the point (X, Y) where the lightness value R(X, Y) is greater than the maximum grayness value V(=O).

Let p and (x) be S, which gives the maximum gray value V, and Q.

The maximum shading value V is determined by setting (Y) to T (W-a), and when this maximum shading value V is larger than the threshold value vTH, there is a white line, and when it is smaller, there is no white line (W-9).

In this way, all the given skeleton constituent points are processed, and the point having the maximum gray value on the parameter plane (P, Q) can be regarded as the optimal straight line.

For example, consider a case where only one white line is visible in the original image, for example, when only the white line at the top right is visible.
G(i,j) extracted as a skeleton line descending to the right does not contain any white line data. Therefore G(i
, j) is expected to be smaller than that when the white line is visible. Therefore, by setting a threshold value to the maximum gradation value V on the parameter plane upon HOUGH conversion, it becomes possible to deal with the situation even when only one white line is visible. In the above example, the maximum gradation value of the parameter plane obtained from the skeleton line G(i, j) descending to the right is less than V, and the white line descending to the right is not detected.

The white line position detected in this way can be regarded as the path end position, and can be used for travel control of an automatic guided vehicle.

Note that the white line i of the two trees obtained by the HOUGH transformation as shown in FIG. 3 (W) and 1. The fulcrum vp of can be regarded as the vanishing point of the i angle measurement image.

(Effects of the Invention) As explained above, in the present invention,! [lNNOn both screens, detect only the area whose left edge is the point where the density changes from dark to bright and the right edge is the point where the density changes from bright to dark, and use the center point of that area as a white line candidate. Since the white line is detected by setting a threshold value to the maximum gradation value of the parameter plane obtained by HOUGH conversion, it is possible to detect the external environment without using markers placed on the driving road or wall surface as in the past. Because it can be recognized, there is no need to reinstall or install new markers even if the factory layout is changed or expanded, making it highly versatile, greatly reducing equipment costs, and reducing the hassle of maintenance management. It is also advantageous in terms of running costs. Furthermore, detection is not affected by slight dirt on the white line, and the position of the white line can be detected stably. Furthermore, there is an advantage that the HOυGH conversion for determining the white line can be performed stably and quickly.

[Brief explanation of the drawing]

Fig. 1 is an overall configuration diagram of a white line detection device according to the present invention;
The figure is a block diagram of an embodiment of the white line detection device according to the present invention, and FIG. 3 shows images obtained in each step of image processing for white line detection according to the present invention.
(B) is an edge constituent point image, (C) is a vertical edge removed image, (D) is a horizontal search range, (E) is a white line candidate image, (F) is a skeleton line image, (G) is an upper right image. Skeleton image, (C) is a skeleton image descending to the right, (S) is an image showing white line detection results,
Figure 4 is a flowchart of white line candidate extraction, Figure 5 is a flowchart of skeleton line image purification, Figure 6 is a skeleton line pattern, Figure 7 is an explanatory diagram of HOUGH conversion, and Figure 8 is a flowchart of HOUGH conversion.
It is a flowchart of UGH conversion. ioo...edge detection means, 200--white line candidate detection means, 300--pattern classification means, 400...
White line detection means patent applicant Nissan Motor Co., Ltd. agent Patent attorney Hiroshi Suzuki

Claims (1)

[Claims] an edge detection means for detecting points where the gray scale value changes from dark to light and points where the gray value changes from bright to dark over the entire image in an observed image including a white line, and detecting an edge formed by a set of detected points; a white line candidate detecting means for detecting an area formed by the area having a predetermined width or less as a white line candidate, and a pattern classification for extracting a center point of the area and classifying a white line pattern from a set of the center points. and white line detection means for detecting white lines by performing HOUGH transformation on the classified patterns.